Embodied Simulation Based on Autobiographical Memory

The ability to generate and exploit internal models of the body, the environment, and their interaction is crucial for survival. Referred to as a forward model, this simulation capability plays an important role in motor control. In this context, the motor command is sent to the forward model in parallel with its actual execution. The results of the actual and simulated execution are then compared, and the consequent error signal is used to correct the movement. Here we demonstrate how the iCub robot can (a) accumulate experience in the generation of action within its Autobiographical memory (ABM), (b) consolidate this experience encoded in the ABM memory to populate a semantic memory whose content can then be used to (c) simulate the results of actions. This simulation can be used as a traditional forward model in the control sense, but it can also be used in more extended time as a mental simulation or mental image that can contribute to higher cognitive function such as planning future actions, or even imagining the mental state of another agent. We present the results of the use of such a mental imagery capability in a forward modeling for motor control task, and a classical mentalizing task. Part of the novelty of this research is that the information that is used to allow the simulation of action is purely acquired from experience. In this sense we can say that the simulation capability is embodied in the sensorimotor experience of the iCub robot.